1 /*
2 * Copyright (c) 1999, 2023, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
24
25 #include "precompiled.hpp"
26 #include "c1/c1_MacroAssembler.hpp"
27 #include "c1/c1_Runtime1.hpp"
28 #include "compiler/compilerDefinitions.inline.hpp"
29 #include "gc/shared/barrierSet.hpp"
30 #include "gc/shared/barrierSetAssembler.hpp"
31 #include "gc/shared/collectedHeap.hpp"
32 #include "gc/shared/tlab_globals.hpp"
33 #include "interpreter/interpreter.hpp"
34 #include "oops/arrayOop.hpp"
35 #include "oops/markWord.hpp"
36 #include "runtime/basicLock.hpp"
37 #include "runtime/os.hpp"
38 #include "runtime/sharedRuntime.hpp"
39 #include "runtime/stubRoutines.hpp"
40
41 int C1_MacroAssembler::lock_object(Register hdr, Register obj, Register disp_hdr, Register tmp, Label& slow_case) {
42 const int aligned_mask = BytesPerWord -1;
43 const int hdr_offset = oopDesc::mark_offset_in_bytes();
44 assert(hdr == rax, "hdr must be rax, for the cmpxchg instruction");
45 assert_different_registers(hdr, obj, disp_hdr, tmp);
46 int null_check_offset = -1;
47
48 verify_oop(obj);
49
50 // save object being locked into the BasicObjectLock
51 movptr(Address(disp_hdr, BasicObjectLock::obj_offset()), obj);
52
53 null_check_offset = offset();
54
55 if (DiagnoseSyncOnValueBasedClasses != 0) {
56 load_klass(hdr, obj, rscratch1);
57 movl(hdr, Address(hdr, Klass::access_flags_offset()));
58 testl(hdr, JVM_ACC_IS_VALUE_BASED_CLASS);
59 jcc(Assembler::notZero, slow_case);
60 }
61
62 // Load object header
63 movptr(hdr, Address(obj, hdr_offset));
64
65 if (LockingMode == LM_LIGHTWEIGHT) {
66 #ifdef _LP64
67 const Register thread = r15_thread;
68 #else
69 const Register thread = disp_hdr;
70 get_thread(thread);
71 #endif
72 fast_lock_impl(obj, hdr, thread, tmp, slow_case);
73 } else if (LockingMode == LM_LEGACY) {
74 Label done;
75 // and mark it as unlocked
76 orptr(hdr, markWord::unlocked_value);
77 // save unlocked object header into the displaced header location on the stack
78 movptr(Address(disp_hdr, 0), hdr);
79 // test if object header is still the same (i.e. unlocked), and if so, store the
80 // displaced header address in the object header - if it is not the same, get the
81 // object header instead
82 MacroAssembler::lock(); // must be immediately before cmpxchg!
83 cmpxchgptr(disp_hdr, Address(obj, hdr_offset));
84 // if the object header was the same, we're done
85 jcc(Assembler::equal, done);
86 // if the object header was not the same, it is now in the hdr register
87 // => test if it is a stack pointer into the same stack (recursive locking), i.e.:
88 //
89 // 1) (hdr & aligned_mask) == 0
90 // 2) rsp <= hdr
91 // 3) hdr <= rsp + page_size
92 //
93 // these 3 tests can be done by evaluating the following expression:
94 //
95 // (hdr - rsp) & (aligned_mask - page_size)
96 //
97 // assuming both the stack pointer and page_size have their least
98 // significant 2 bits cleared and page_size is a power of 2
99 subptr(hdr, rsp);
100 andptr(hdr, aligned_mask - (int)os::vm_page_size());
101 // for recursive locking, the result is zero => save it in the displaced header
102 // location (null in the displaced hdr location indicates recursive locking)
103 movptr(Address(disp_hdr, 0), hdr);
104 // otherwise we don't care about the result and handle locking via runtime call
105 jcc(Assembler::notZero, slow_case);
106 // done
107 bind(done);
108 }
109
110 inc_held_monitor_count();
111
112 return null_check_offset;
113 }
114
115 void C1_MacroAssembler::unlock_object(Register hdr, Register obj, Register disp_hdr, Label& slow_case) {
116 const int aligned_mask = BytesPerWord -1;
117 const int hdr_offset = oopDesc::mark_offset_in_bytes();
118 assert(disp_hdr == rax, "disp_hdr must be rax, for the cmpxchg instruction");
119 assert(hdr != obj && hdr != disp_hdr && obj != disp_hdr, "registers must be different");
120 Label done;
121
122 if (LockingMode != LM_LIGHTWEIGHT) {
123 // load displaced header
124 movptr(hdr, Address(disp_hdr, 0));
125 // if the loaded hdr is null we had recursive locking
126 testptr(hdr, hdr);
127 // if we had recursive locking, we are done
128 jcc(Assembler::zero, done);
129 }
130
131 // load object
132 movptr(obj, Address(disp_hdr, BasicObjectLock::obj_offset()));
133 verify_oop(obj);
134
135 if (LockingMode == LM_LIGHTWEIGHT) {
136 movptr(disp_hdr, Address(obj, hdr_offset));
137 andptr(disp_hdr, ~(int32_t)markWord::lock_mask_in_place);
138 fast_unlock_impl(obj, disp_hdr, hdr, slow_case);
139 } else if (LockingMode == LM_LEGACY) {
140 // test if object header is pointing to the displaced header, and if so, restore
141 // the displaced header in the object - if the object header is not pointing to
142 // the displaced header, get the object header instead
143 MacroAssembler::lock(); // must be immediately before cmpxchg!
144 cmpxchgptr(hdr, Address(obj, hdr_offset));
145 // if the object header was not pointing to the displaced header,
146 // we do unlocking via runtime call
147 jcc(Assembler::notEqual, slow_case);
148 // done
149 }
150 bind(done);
151 dec_held_monitor_count();
152 }
153
154
155 // Defines obj, preserves var_size_in_bytes
156 void C1_MacroAssembler::try_allocate(Register obj, Register var_size_in_bytes, int con_size_in_bytes, Register t1, Register t2, Label& slow_case) {
157 if (UseTLAB) {
158 tlab_allocate(noreg, obj, var_size_in_bytes, con_size_in_bytes, t1, t2, slow_case);
159 } else {
160 jmp(slow_case);
161 }
162 }
163
164
165 void C1_MacroAssembler::initialize_header(Register obj, Register klass, Register len, Register t1, Register t2) {
166 assert_different_registers(obj, klass, len);
167 movptr(Address(obj, oopDesc::mark_offset_in_bytes()), checked_cast<int32_t>(markWord::prototype().value()));
168 #ifdef _LP64
169 if (UseCompressedClassPointers) { // Take care not to kill klass
170 movptr(t1, klass);
171 encode_klass_not_null(t1, rscratch1);
172 movl(Address(obj, oopDesc::klass_offset_in_bytes()), t1);
173 } else
174 #endif
175 {
176 movptr(Address(obj, oopDesc::klass_offset_in_bytes()), klass);
177 }
178
179 if (len->is_valid()) {
180 movl(Address(obj, arrayOopDesc::length_offset_in_bytes()), len);
181 }
182 #ifdef _LP64
183 else if (UseCompressedClassPointers) {
184 xorptr(t1, t1);
185 store_klass_gap(obj, t1);
186 }
187 #endif
188 }
189
190
191 // preserves obj, destroys len_in_bytes
192 void C1_MacroAssembler::initialize_body(Register obj, Register len_in_bytes, int hdr_size_in_bytes, Register t1) {
193 assert(hdr_size_in_bytes >= 0, "header size must be positive or 0");
194 Label done;
195
196 // len_in_bytes is positive and ptr sized
197 subptr(len_in_bytes, hdr_size_in_bytes);
198 zero_memory(obj, len_in_bytes, hdr_size_in_bytes, t1);
199 bind(done);
200 }
201
202
203 void C1_MacroAssembler::allocate_object(Register obj, Register t1, Register t2, int header_size, int object_size, Register klass, Label& slow_case) {
204 assert(obj == rax, "obj must be in rax, for cmpxchg");
205 assert_different_registers(obj, t1, t2); // XXX really?
206 assert(header_size >= 0 && object_size >= header_size, "illegal sizes");
207
208 try_allocate(obj, noreg, object_size * BytesPerWord, t1, t2, slow_case);
209
210 initialize_object(obj, klass, noreg, object_size * HeapWordSize, t1, t2, UseTLAB);
211 }
212
213 void C1_MacroAssembler::initialize_object(Register obj, Register klass, Register var_size_in_bytes, int con_size_in_bytes, Register t1, Register t2, bool is_tlab_allocated) {
214 assert((con_size_in_bytes & MinObjAlignmentInBytesMask) == 0,
215 "con_size_in_bytes is not multiple of alignment");
216 const int hdr_size_in_bytes = instanceOopDesc::header_size() * HeapWordSize;
217
218 initialize_header(obj, klass, noreg, t1, t2);
219
220 if (!(UseTLAB && ZeroTLAB && is_tlab_allocated)) {
221 // clear rest of allocated space
222 const Register t1_zero = t1;
223 const Register index = t2;
224 const int threshold = 6 * BytesPerWord; // approximate break even point for code size (see comments below)
225 if (var_size_in_bytes != noreg) {
226 mov(index, var_size_in_bytes);
227 initialize_body(obj, index, hdr_size_in_bytes, t1_zero);
228 } else if (con_size_in_bytes <= threshold) {
229 // use explicit null stores
230 // code size = 2 + 3*n bytes (n = number of fields to clear)
231 xorptr(t1_zero, t1_zero); // use t1_zero reg to clear memory (shorter code)
232 for (int i = hdr_size_in_bytes; i < con_size_in_bytes; i += BytesPerWord)
233 movptr(Address(obj, i), t1_zero);
234 } else if (con_size_in_bytes > hdr_size_in_bytes) {
235 // use loop to null out the fields
236 // code size = 16 bytes for even n (n = number of fields to clear)
237 // initialize last object field first if odd number of fields
238 xorptr(t1_zero, t1_zero); // use t1_zero reg to clear memory (shorter code)
239 movptr(index, (con_size_in_bytes - hdr_size_in_bytes) >> 3);
240 // initialize last object field if constant size is odd
241 if (((con_size_in_bytes - hdr_size_in_bytes) & 4) != 0)
242 movptr(Address(obj, con_size_in_bytes - (1*BytesPerWord)), t1_zero);
243 // initialize remaining object fields: rdx is a multiple of 2
244 { Label loop;
245 bind(loop);
246 movptr(Address(obj, index, Address::times_8, hdr_size_in_bytes - (1*BytesPerWord)),
247 t1_zero);
248 NOT_LP64(movptr(Address(obj, index, Address::times_8, hdr_size_in_bytes - (2*BytesPerWord)),
249 t1_zero);)
250 decrement(index);
251 jcc(Assembler::notZero, loop);
252 }
253 }
254 }
255
256 if (CURRENT_ENV->dtrace_alloc_probes()) {
257 assert(obj == rax, "must be");
258 call(RuntimeAddress(Runtime1::entry_for(Runtime1::dtrace_object_alloc_id)));
259 }
260
261 verify_oop(obj);
262 }
263
264 void C1_MacroAssembler::allocate_array(Register obj, Register len, Register t1, Register t2, int header_size, Address::ScaleFactor f, Register klass, Label& slow_case) {
265 assert(obj == rax, "obj must be in rax, for cmpxchg");
266 assert_different_registers(obj, len, t1, t2, klass);
267
268 // determine alignment mask
269 assert(!(BytesPerWord & 1), "must be a multiple of 2 for masking code to work");
270
271 // check for negative or excessive length
272 cmpptr(len, checked_cast<int32_t>(max_array_allocation_length));
273 jcc(Assembler::above, slow_case);
274
275 const Register arr_size = t2; // okay to be the same
276 // align object end
277 movptr(arr_size, header_size * BytesPerWord + MinObjAlignmentInBytesMask);
278 lea(arr_size, Address(arr_size, len, f));
279 andptr(arr_size, ~MinObjAlignmentInBytesMask);
280
281 try_allocate(obj, arr_size, 0, t1, t2, slow_case);
282
283 initialize_header(obj, klass, len, t1, t2);
284
285 // clear rest of allocated space
286 const Register len_zero = len;
287 initialize_body(obj, arr_size, header_size * BytesPerWord, len_zero);
288
289 if (CURRENT_ENV->dtrace_alloc_probes()) {
290 assert(obj == rax, "must be");
291 call(RuntimeAddress(Runtime1::entry_for(Runtime1::dtrace_object_alloc_id)));
292 }
293
294 verify_oop(obj);
295 }
296
297
298
299 void C1_MacroAssembler::inline_cache_check(Register receiver, Register iCache) {
300 verify_oop(receiver);
301 // explicit null check not needed since load from [klass_offset] causes a trap
302 // check against inline cache
303 assert(!MacroAssembler::needs_explicit_null_check(oopDesc::klass_offset_in_bytes()), "must add explicit null check");
304 int start_offset = offset();
305
306 if (UseCompressedClassPointers) {
307 load_klass(rscratch1, receiver, rscratch2);
308 cmpptr(rscratch1, iCache);
309 } else {
310 cmpptr(iCache, Address(receiver, oopDesc::klass_offset_in_bytes()));
311 }
312 // if icache check fails, then jump to runtime routine
313 // Note: RECEIVER must still contain the receiver!
314 jump_cc(Assembler::notEqual,
315 RuntimeAddress(SharedRuntime::get_ic_miss_stub()));
316 const int ic_cmp_size = LP64_ONLY(10) NOT_LP64(9);
317 assert(UseCompressedClassPointers || offset() - start_offset == ic_cmp_size, "check alignment in emit_method_entry");
318 }
319
320
321 void C1_MacroAssembler::build_frame(int frame_size_in_bytes, int bang_size_in_bytes) {
322 assert(bang_size_in_bytes >= frame_size_in_bytes, "stack bang size incorrect");
323 // Make sure there is enough stack space for this method's activation.
324 // Note that we do this before doing an enter(). This matches the
325 // ordering of C2's stack overflow check / rsp decrement and allows
326 // the SharedRuntime stack overflow handling to be consistent
327 // between the two compilers.
328 generate_stack_overflow_check(bang_size_in_bytes);
329
330 push(rbp);
331 if (PreserveFramePointer) {
332 mov(rbp, rsp);
333 }
334 #if !defined(_LP64) && defined(COMPILER2)
335 if (UseSSE < 2 && !CompilerConfig::is_c1_only_no_jvmci()) {
336 // c2 leaves fpu stack dirty. Clean it on entry
337 empty_FPU_stack();
338 }
339 #endif // !_LP64 && COMPILER2
340 decrement(rsp, frame_size_in_bytes); // does not emit code for frame_size == 0
341
342 BarrierSetAssembler* bs = BarrierSet::barrier_set()->barrier_set_assembler();
343 // C1 code is not hot enough to micro optimize the nmethod entry barrier with an out-of-line stub
344 bs->nmethod_entry_barrier(this, nullptr /* slow_path */, nullptr /* continuation */);
345 }
346
347
348 void C1_MacroAssembler::remove_frame(int frame_size_in_bytes) {
349 increment(rsp, frame_size_in_bytes); // Does not emit code for frame_size == 0
350 pop(rbp);
351 }
352
353
354 void C1_MacroAssembler::verified_entry(bool breakAtEntry) {
355 if (breakAtEntry || VerifyFPU) {
356 // Verified Entry first instruction should be 5 bytes long for correct
357 // patching by patch_verified_entry().
358 //
359 // Breakpoint and VerifyFPU have one byte first instruction.
360 // Also first instruction will be one byte "push(rbp)" if stack banging
361 // code is not generated (see build_frame() above).
362 // For all these cases generate long instruction first.
363 fat_nop();
364 }
365 if (breakAtEntry) int3();
366 // build frame
367 IA32_ONLY( verify_FPU(0, "method_entry"); )
368 }
369
370 void C1_MacroAssembler::load_parameter(int offset_in_words, Register reg) {
371 // rbp, + 0: link
372 // + 1: return address
373 // + 2: argument with offset 0
374 // + 3: argument with offset 1
375 // + 4: ...
376
377 movptr(reg, Address(rbp, (offset_in_words + 2) * BytesPerWord));
378 }
379
380 #ifndef PRODUCT
381
382 void C1_MacroAssembler::verify_stack_oop(int stack_offset) {
383 if (!VerifyOops) return;
384 verify_oop_addr(Address(rsp, stack_offset));
385 }
386
387 void C1_MacroAssembler::verify_not_null_oop(Register r) {
388 if (!VerifyOops) return;
389 Label not_null;
390 testptr(r, r);
391 jcc(Assembler::notZero, not_null);
392 stop("non-null oop required");
393 bind(not_null);
394 verify_oop(r);
395 }
396
397 void C1_MacroAssembler::invalidate_registers(bool inv_rax, bool inv_rbx, bool inv_rcx, bool inv_rdx, bool inv_rsi, bool inv_rdi) {
398 #ifdef ASSERT
399 if (inv_rax) movptr(rax, 0xDEAD);
400 if (inv_rbx) movptr(rbx, 0xDEAD);
401 if (inv_rcx) movptr(rcx, 0xDEAD);
402 if (inv_rdx) movptr(rdx, 0xDEAD);
403 if (inv_rsi) movptr(rsi, 0xDEAD);
404 if (inv_rdi) movptr(rdi, 0xDEAD);
405 #endif
406 }
407
408 #endif // ifndef PRODUCT